Seven Signals That Will Reveal AI’s True Economic Impact

Aggregate Employment And Wages

So far we have seen little impact on total employment and wage rates from AI. The economy is large and always changing, so small impacts are hard to see. And since AI has become widely used, other changes have suddenly impacted the labor market: such as immigration policy, tariffs and the Iran War. Longer-term issues are also at work, including baby boomer retirement, the huge rise of healthcare employment, DEI practices and skills gaps. With so many other changes going on, AI impacts would have to be large to show up in the data. Small or even moderate shifts could easily be missed by the most careful analysis.

However, this is the top issue to monitor on an on-going basis.

AI Investment Before Payoff

Many benefits of new technology require initial investments in training, equipment or process restructuring. An individual worker using a chatbot will benefit from trying different types of questions and prompts. The person will learn by trial which types of problems the AI helps with quickly and which are a waste of time. The time spent learning is like a capital expenditure: pay now for future benefit.

More significantly, the greatest benefits of a new technology often come when business processes have been adjusted to the tool’s capabilities. In the days of belt-driven factories, machines crowded around the power shaft, leaving little room to place equipment where workflow made the most sense. Electrification meant that factory layouts could be more flexible, leading to the assembly line. But it took years and effort to reorganize factories to benefit from the new technology.

AI will enable some tasks to be done more easily, but taking full advantage of the new capabilities will require different data collection and information transfer paths. This will take time and entail some up-front costs. This issue will not be obvious from the usual economic data; we’ll need to watch individual business actions to see it progress.

Business leaders address this issue in their own organizations as well as watching it in broad economic data. Investments needed to exploit AI should be evaluated as if they are capital expenditures, though some will show up in financial reports as ordinary expenses. I highlighted business strategy regarding this in an earlier article.

Productivity Effects In The Data

Aggregate productivity—output per hour worked—is critical to overall economic growth. If it’s high, then fiscal and monetary policy can stimulate the economy. But if productivity growth is low, then stimulus just turns into inflation—too many dollars chasing too few goods. To get productivity data right, though, inflation data has to be accurate. If the price of a car goes up, how much of the price change is due to higher quality (more safety sensors, more comfortable seats) and how much due to inflation? That is hard to figure out given the huge number of different products being produced, the wide variety of features and the continual change in many products.

Some of AI’s impacts will be to boost quality in ways that may not be measured. Thus productivity is actually higher—because products are better in some way. If my credit card company uses AI to answer my account questions faster and more accurately, I have a better product—but the economic statistics don’t show that.

AI may very well help develop better statistics, but right now we’ll watch the existing figures and hope to see much greater output per hour worked. But we’ll look for anecdotal evidence of quality improvement.

Augmentation Or Automation

Studies of particular workers and the tasks they perform mostly show AI as augmenting workers, enabling them to perform faster and better. This is particularly true of less-experienced workers. We have seen little replacement of workers—automation—yet. This will certainly happen in some cases, and probably many cases. First we’ll see it anecdotally, then see it in the aggregate employment and wage data.

Business leaders in different sectors may respond differently. Companies facing severe cost competition will want to cut staffing whenever possible. Other organizations may have a challenge with employee retention stemming from overwork and burnout. That seems typical of healthcare today. Helping staff members finish their work with less time and stress may be better the better strategy.

AI Diffusion Across Industries And Companies

The adoption of AI varies widely today, according to surveys of companies and sectors of the economy. Competitive pressure will lead companies to either take up productivity tools or fall behind. In fact, the historical data show that the most innovative companies with respect to labor-saving technology added workers—at the expense of their competitors who did not use cost-reduction tools.

Business leaders will want to watch their own sectors to assess how much their competitors are benefitting from AI. They may also want to look broadly at other sectors for implementation ideas. I wrote an article summarizing some lessons from a McKinsey survey, and more recent surveys are available. There’s always risk of being on the “bleeding edge,” spending money for little gain. But with technology changing rapidly, no one can ignore the risk of becoming uncompetitive due to failure to implement cost-saving technology.

Inflation Dynamics From AI

Inflation affects consumers and businesses and also the decisions that the Federal Reserve makes on interest rates. The effect of AI on inflation depends critically on general expectations. If decision-makers expect big future gains in productivity from AI, they will be willing to spend a great deal to implement AI. In addition to implementation expenses, some consumers may also anticipate a stronger economy and spend more today. That will be inflationary until total production actually increases.

But if productivity gains are not expected, current spending does not increase much. When productivity eventually shows up, the increased supply of goods and services will tend to push prices down.

In my previous writings, I ignored this point, instead emphasizing that monetary policy needs to align with productivity growth. That’s still relevant, but expectations are also important.

Entry-Level Hiring And AI

Some occupations may be seeing less entry-level hiring. The data are soft but align with a reasonable hypothesis: Companies still need seasoned professionals, but AI can replace many of the lower-level positions. For example, in years past a senior programmer would assign some simple, tedious programming tasks to new workers. Now an AI coding agent handles the work.

Lack of entry level hiring has impacts beyond the total employment numbers. How does one become a senior programmer without the experience of being a junior programmer working with an experienced person?

The flip side, though, is skill compression. In some occupations, less experienced people with AI may be able to do the job of seasoned workers. AI has helped customer service employees more at the low end than the high end of their experience. For example, the experienced representative knew from past cases that when the customer had three particular characteristics, the usual answer would be wrong. The rookie wouldn’t know that—but the AI knows it. The rookie with AI performs as well as the veteran.

This is an issue that should be closely watched in occupations where experience has proved useful in the past.

AI And The Economy: Conclusion

The broad conclusions that many of us economists developed early in the AI space still stand. Productivity will increase. As a result, goods and services prices would fall relative to wages, so that general purchasing power of consumers would improve. But there will be people who become worse off due to job losses, despite general improvement for most people.

The details of timing and who gains the most and who is hurt the most are evolving; these topics will help us see the changes.

An excellent background paper by Eric Fruits and Kristian Stout explains in more detail the issues discussed above.